Computer systems use memory devices, such as dynamic random access memory (“DRAM”) devices, to store data that are accessed by a processor. These memory devices are normally used as system memory in a computer system. In a typical computer system, the processor communicates with the system memory through a processor bus and a memory controller. The memory devices of the system memory, typically arranged in memory modules having multiple memory devices, are coupled through a memory bus to the memory controller. The processor issues a memory request, which includes a memory command, such as a read command, and an address designating the location from which data or instructions are to be read. The memory controller uses the command and address to generate appropriate command signals as well as row and column addresses, which are applied to the system memory through the memory bus. In response to the commands and addresses, data are transferred between the system memory and the processor. The memory controller is often part of a system controller, which also includes bus bridge circuitry for coupling the processor bus to an expansion bus, such as a PCI bus.
In memory systems, high data bandwidth is desirable. Generally, bandwidth limitations are not related to the memory controllers since the memory controllers sequence data to and from the system memory as fast as the memory devices allow. One approach that has been taken to increase bandwidth is to increase the speed of the memory data bus coupling the memory controller to the memory devices. Thus, the same amount of information can be moved over the memory data bus in less time. However, despite increasing memory data bus speeds, a corresponding increase in bandwidth does not result. One reason for the non-linear relationship between data bus speed and bandwidth is the hardware limitations within the memory devices themselves. That is, the memory controller has to schedule all memory commands to the memory devices such that the hardware limitations are honored. Although these hardware limitations can be reduced to some degree through the design of the memory device, a compromise must be made because reducing the hardware limitations typically adds cost, power, and/or size to the memory devices, all of which are undesirable alternatives. Thus, given these constraints, although h is easy for memory devices to move “well-behaved” traffic at ever increasing rates, for example, sequel traffic to the same page of a memory device, it is much more difficult for the memory devices to resolve “badly-behaved traffic,” such as bouncing between different pages or banks of the memory device. As a result, the increase in memory data bus bandwidth does not yield a corresponding increase in information bandwidth.
In addition to the limited bandwidth between processors and memory devices, the performance of computer systems is also limited by latency problems that increase the time required to read data from system memory devices. More specifically, when a memory device read command is coupled to a system memory device, such as a synchronous DRAM (“SDRAM”) device, the read data are output from the SDRAM device only after a delay of several clock periods. Therefore, although SDRAM devices can synchronously output burst data at a high data rate, the delay in initially providing the data can significantly slow the operating speed of a computer system using such SDRAM devices. Increasing the memory data bus speed can be used to help alleviate the latency issue. However, as with bandwidth, the increase in memory data bus speeds do not yield a linear reduction of latency, for essentially the same reasons previously discussed.
Although increasing memory data bus speed has, to some degree, been successful in increasing bandwidth and reducing latency, other issues are raised by this approach. For example, as the speed of the memory data bus increases, loading on the memory bus needs to be decreased in order to maintain signal integrity since traditionally, there has only been wire between the memory controller and the memory slots into which the memory modules are plugged. Several approaches have been taken to accommodate the increase in memory data bus speed. For example, reducing the number of memory slots, adding buffer circuits on a memory module in order to provide sufficient fanout of control signals to the memory devices on the memory module, and providing multiple memory device interfaces on the memory module since there are too few memory module connectors on a single memory device interface. The effectiveness of these conventional approaches are, however, limited. A reason why these techniques were used in the past is that it was cost-effective to do so. However, when only one memory module can be plugged in per interface, it becomes too costly to add a separate memory interface for each required memory slot. In other words, it pushes the system controllers package out of the commodity range and into the boutique range, thereby, greatly adding cost.
One recent approach that allows for increased memory data bus speed in a cost effective manner is the use of multiple memory devices coupled to the processor through a memory hub. In a memory hub architecture, or a hub-based memory sub-system, a system controller or memory controller is coupled over a high speed bi-directional or unidirectional memory controller/hub interface to several memory modules. Typically, the memory modules axe coupled in a point-to-point or daisy chain architecture such that the memory modules are connected one to another in series. Thus, the memory controller is coupled to a first memory module, with the first memory module connected to a second memory module, and the second memory module coupled to a third memory module, and so on in a daisy chain fashion.
Each memory module includes a memory hub that is coupled to the memory controller/hub interface and a number of memory devices on the module, with the memory hubs efficiently routing memory requests and responses between the controller and the memory devices over the memory controller/hub interface. Computer systems employing this architecture can use a high-speed memory data bus since signal integrity can be maintained on the memory data bus. Moreover, this architecture also provides for easy expansion of the system memory without concern for degradation in signal quality as more memory modules are added, such as occurs in conventional memory bus architectures.
Although computer systems using memory hubs can provide superior performance, various factors may affect the performance of the memory system. For example, the manner in which the flow of read data upstream (i.e., back to the memory hob controller in the computer system) from one memory hub to another is managed will affect read latency. The management of the flow of read data by a memory hub may be generally referred to as arbitration, with each memory hub arbitrating between local memory read responses and upstream memory read responses. That is, each memory hub determines whether to send local memory read responses first or to forward memory read responses from downstream (i.e., further away from the memory hub controller) memory hubs first. Although the determination of which memory read response has lower priority will only affect the latency of that specific memory read response, the additive effect of the memory read responses having increased latency will affect the overall latency of the memory system. Consequently, the arbitration technique employed by a memory hub directly affects the performance of the overall memory system. Additionally, the implementation of the arbitration scheme will affect the overall read latency as well, since inefficient implementation will negatively impact system memory performance despite utilizing a desirable arbitration scheme. Therefore, there is a need for a system and method for implementing an arbitration scheme for managing memory responses in a system memory having a memory hub architecture.